how monetary policy shaped the housing...

How Monetary PolicyShaped the Housing Boom

Itamar Drechsler1 Alexi Savov2 Philipp Schnabl3

1Wharton and NBER 2NYU Stern and NBER 3NYU Stern, CEPR, and NBER

May 2019

Monetary Policy and the Housing Boom

1. The role of monetary policy in the housing boom remains unresolved

- on one side: Taylor (2007) argues that the Fed kept rates “too lowfor too long,” leading to excessive investment in housing

- on the other side: Bernanke (2010) argues that monetary policy wasnot too loose. Real culprit was a decline in mortgage lendingstandards that accompanied the shift from traditional bank portfoliolending to securitized lending

2. This debate is unresolved in part because the housing boom actuallyaccelerated from 2003 to 2006, when the Fed tightened by 425 bps

- mortgage spreads narrowed in mid-2003 (Justiniano et al., 2017)

- lending standards fell and house prices took off

⇒ What impact, if any, did Fed tightening have on the housingboom?

Drechsler, Savov, and Schnabl (2018)

Mortgage lending and the housing boom

1. Expansion of mortgage lending was a key driver of the housing boom(e.g., Mian and Sufi, 2009)

2. Private-Label Securitization (PLS) and non-bank lending grewdisproportionately relative to bank portfolio lending and GSEs

- areas with more non-banks experienced a bigger housing boom(Mian and Sufi, 2018)

3. Relation to monetary policy?“The deposits channel” (Drechsler, Savov, and Schnabl, 2017)

→ as the Fed tightens, bank deposits flow out

→ banks contract their portfolio lending

→ lending shifts to PLS and non-banks?


In this paper we find

1. Fed tightening led to large outflows of bank deposits, as predicted bythe deposits channel

2. This induced a substantial contraction in bank portfolio mortgagelending

3. But, it also induced a large shift to PLS, led by non-banks, whichlargely offset the contraction in bank portfolio lending

- rate of substitution: 65% of reduced bank portfolio lending cameback as PLS (most by non-banks)

- mortgage market shifted from stable deposit funding to run-pronewholesale funding

⇒ Fed tightening:

- was ineffective at curbing mortgage lending

- accelerated the shift to PLS/non-banks

- raised exposure of housing market to runs/freezes


Related literature

1. Mortgage lending, housing booms, and financial crises: Mian and Sufi (2009);Adelino, Schoar, and Severino (2016); Schularick and Taylor (2012), Jorda,Schularick, and Taylor (2016); Justiniano, Primiceri and Tambalotti (2017)

2. Bank lending/deposits channel of monetary policy: Bernanke (1983); Bernankeand Blinder (1988); Kashyap and Stein (1994, 2000); Landier, Sraer, andThesmar (2013); Scharfstein and Sunderam (2016); Hanson, Shleifer, Stein, andVishny (2015); Drechsler, Savov and Schnabl (2017)

3. Monetary policy and financial stability: Kashyap, Stein, and Wilcox (1993);Stein (1998, 2012); Diamond and Rajan (2012); Greenwood, Hanson, and Stein(2014); Stein and Sunderam (2016); Drechsler, Savov and Schnabl (2018); Xiao(2018)

4. Competition between banks and shadow banks: Gennaioli, Shleifer, and Vishny(2013); Sunderam (2014); Moreira and Savov (2017); Xiao (2018); Buchak,Matvos, Piskorski, and Seru (2018)


Private-label securitization (PLS) and Monetary Policy

0%

1%

2%

3%

4%

5%

6%

7%

0%

10%

20%

30%

40%

50%

60%

70%

19

96

19

97

19

98

19

99

20

00

20

01

20

02

20

03

20

04

20

05

20

06

20

07

20

08

20

09

20

10

20

11

20

12

20

13

20

14

20

15

20

16

20

17

20

18

PLS share of securitized issuance 1-Year Treasury rate (right axis)

1. Strong positive co-movement between interest rates and PLS since 2002

- before 2002, PLS share of total securitization was < 25%

- mid-2003 to 2006: as Fed tightens, PLS share rises sharply to ≈ 60%

- PLS non-existent during ZLB period

- has re-emerged as interest rates riseDrechsler, Savov, and Schnabl (2018)

The deposits channel (DSS 2017)

0%

2%

4%

6%

8%

10%

1987 1990 1993 1996 1999 2002 2005 2008

Core deposit rate Fed funds rate

-6%

-4%

-2%

0%

2%

4%

-4%

0%

4%

8%

12%

16%

1987 1990 1993 1996 1999 2002 2005 2008

Core deposit growth Δ Fed funds rate (right axis)

1. Fed tightening induces outflows of bank deposits

- banks have market power over retail (core) deposit markets

- when the Fed funds rate rises, banks charge higher deposit spreads

- this causes deposits to flow out

2. Deposits are the main source of bank funding (77% of liabilities)/Banks value deposits for their unique stability

⇒ deposit outflows induce banks to contract lending


The deposits channel, 2003–2006

1. Did Fed tightening shrink deposit supply during the housing boom?

- identification challenge: Fed tightening also weakens loan demand

2. Cross-sectional analysis: deposit spreads should rise more anddeposits should flow out more in less competitive areas

- measure local competition using deposit spread betas:for all branches b in county c, run

∆DepositSpreadb,c,t = βc∆FedFundst + εb,c,t

- βc captures pricing power of branches in county c (Branch beta)- estimate βc ’s from prior cycles (pre-2002)

3. Control for loan demand by comparing branches of the same bank(“within-bank estimation”)

- identifying assumption: a deposit dollar raised at one branch can belent out at another branch


Branch-level analysis

Data:

1. Branch- and product-level deposit rates: Ratewatch (1997–2015)

2. Branch-level deposits: FDIC (1994–2015)

3. Bank balance sheets: U.S. Call Reports (1986–2015)

4. County characteristics: County Business Patterns

Measures:

1. Deposit spread = Fed funds rate − deposits rate

2. Branch betas: estimate using pre-2002 data, use to predict depositsupply during 2003–2006


Distribution of Branch betas

Branch betas

1. Branch betas average 0.58 ⇒ deposit spreads increase on average by58 bps per 100 bps increase in the Fed funds rate

2. There is substantial cross-sectional variation- DSS (2017) show that local deposit market power is explained by

market concentration, income, education, demographics


Deposit spreads, 2003–2006

∆DepositSpreadbranch,2003−2006 = α + γBranchBeta2002 + ε

Bin-scatter plots:

∆ Savings deposits spread ∆ Small time deposits spread

33.

23.

43.

63.

8S

avin

gs D

epos

it S

prea

d

.4 .5 .6 .7 .8Spread Beta

Coef. = 1.725

1.5

1.6

1.7

1.8

1.9

2S

mal

l Tim

e D

epos

it S

prea

d

.4 .5 .6 .7 .8Spread Beta

Coef. = 1.065

1. Deposit spreads rose strongly during the 2003-2006 period

2. Pre-2002 branch betas strongly predict the deposit spread changes


Deposit growth, 2003–2006

∆Log(Deposits)branch,2003−2006 = α + γBranchBeta2002 + ε

1520

2530

Dep

osit

Gro

wth

.4 .5 .6 .7 .8Branch Beta

Coef. = -32.231

1. Higher branch beta⇒ spread increases more⇒ lower deposit growth

⇒ Fed tightening induces inward shift in deposit supply (higher prices,lower quantities)


Deposit growth, 2003–2006, within-bank estimation

∆Log(Deposits)branch,2003−2006 = µbank + γBranchBeta2002 + ε

Panel B: Deposit Growth(1) (2)

Branch beta −0.322*** −0.213***(5.046) (6.037)

Bank Fixed Effects N YObservations 59,700 57,497R2 0.002 0.186

1. Pre-2002 branch betas predict 2003–2006 deposit growth across differentbranches of the same bank

⇒ not driven by differences in loan demand

⇒ Fed tightening shrank aggregate deposits by 12.4%

- = −0.213× 0.58 (average branch beta)- consistent with aggregate time series


Bank-level analysis

1. Verify branch-level deposits results aggregate up to bank level

2. Extend analysis to asset side of bank balance sheets

3. U.S. Call Reports 1986–2015 (6,356 banks)

- measure deposit market power of bank B using its Bank beta βB :

∆DepositSpreadB,t = αB + βB∆FedFundst + εB,t

- estimate βB (Bank beta) using pre-2002 data

- Bank beta captures a bank’s exposure to the deposits channel

- use Bank betas to predict deposit supply and bank assets during2003–2006


Bank-level deposit supply, 2003–2006

∆yBank,2003−2006 = α + γBankBeta2002 + ε

∆ Core Deposit Spread ∆Log(Core deposits)

2.5

33.

5C

hang

e in

Dep

osit

Spr

ead

.4 .5 .6 .7 .8Bank Beta

Coef. = 2.695

.1.2

.3.4

Dep

osit

Gro

wth

.4 .5 .6 .7 .8Bank Beta

Coef. = -0.550

⇒ Pre-2002 Bank betas predict deposit spreads and deposit growthduring the housing boom

- verifies branch-level results at the bank level (different datasets)


Bank-level real estate loans and securities


∆Log(Assets) ∆ Log(Real Estate Loans)

.15

.2.2

5.3

.35

.4A

sset

Gro

wth

.4 .5 .6 .7 .8Bank Beta

Coef. = -0.541

.3.3

5.4

.45

.5.5

5R

eal E

stat

e Lo

an G

row

th

.4 .5 .6 .7 .8Bank Beta

Coef. = -0.475

⇒ Fed tightening induced a substantial contraction in banks’ holdingsof real estate loans and securities through the deposits channel


Bank-level deposits, real estate loans, and securities


∆ Deposits ∆ Real Estate Loans(1) (2)

Bank Beta −0.262*** −0.213***(0.037) (0.052)

Observations 6,396 6,367R-squared 0.137 0.054

1. Deposits contract by 26% and and real estate loans by 21% at abank with a beta of 1 (maximally exposed) relative to a bank with abeta of 0 (unexposed)

- average bank beta is 0.62 ⇒ implied aggregate impact is 16.2%contraction in deposits, 13.2% contraction in real estate loans


County-level analysis

1. Examine how Fed tightening impacted the level and composition ofmortgage lending through the deposits channel

2. Construct county-level exposure to deposits channel

= average Bank beta in a county, weighted by 2002 portfolio lendingshares:

CountyBetac =∑b

sb,cBankBetab

- county beta mean of 0.53; st. dev. of 0.06

3. Use County betas to predict mortgage lending during the housingboom, 2003–2006

- Focus on bank portfolio and PLS-funding loans: financed privately,either held in banks’ portfolios or sold through PLS → exposed todeposits channel (use GSE loan growth as control)


County-level analysis: empirical strategy

1. Identification challenge: local exposure to deposits channelcorrelated with loan demand over 2003–2006

2. Use county and market structure characteristics as controls

- county: lending, employment, income in 2002- market structure: top-4 lender share (Scharfstein and Sunderam

2016), 2002 PLS share (Mian and Sufi 2018), deposit-weightedcounty beta (uses deposit weights to construct beta)

3. Control for proxies of loan demand

- ∆ income, employment over 2003–2006- ∆ GSE lending over 2003-2006 (since GSE segment is not exposed

to deposits channel)

4. Look at change in PLS and non-bank share over 2003-2006 -controls for total loan demand by scaling by total lending


Bank portfolio lending, 2003–2006

∆Log(Bank portfolio lending)county,2003−2006 = α + γCountyBeta + ε

∆ Bank portfolio lending

0.0

5.1

.15

.2

.4 .5 .6 .7County Beta

Slope = -0.709 (0.151)

⇒ As Fed tightened, counties more exposed to deposits channel saw lessbank portfolio mortgage lending


Bank portfolio lending, 2003–2006

∆ycounty,2003−2006 = α + γCountyBeta + δXcounty + ε

∆ Bank portfolio lending

(3) (4)

County beta −0.436*** −0.486***(0.162) (0.163)

County controls Y Y∆Demand controls N YObs. 2,998 2,750R2 0.138 0.176

1. Portfolio lending is 48.6% lower in a county with beta of 1 (maximallyexposed) than in a county with beta of 0 (unexposed)

⇒ Aggregate reduction due to deposits channel: −0.486× 0.532 = −25.9%

2. Robust to controls for characteristics (lending, employment, income),market structure (local deposit-weighted beta, PLS share, top-4 lendershare), loan demand 2003–06 (∆GSE lending, ∆Income, ∆Employment)


Change in PLS share, 2003–2006

1. Look at market share to control for total loan demand

∆PLS sharecounty,2003−2006 = α + γCountyBeta + ε

Change in PLS lending share

.08

.1.1

2.1

4.1

6


Slope = 0.213 (0.043)

2. As Fed tightens and bank portfolio mortgage lending contracts → marketshifts strongly towards private-label securitization

- intercept ≈ 0 → no growth in PLS share in unexposed counties


Change in PLS share, 2003–2006


∆ PLS lending share

(1) (2)

County beta 0.141*** 0.192***(0.046) (0.043)


1. PLS lending share rises by 19.2% in a county with beta of 1 (maximallyexposed) relative to a county with beta of 0 (unexposed)

2. Aggregate impact: deposits channel can account for a 10.2% increase inPLS share vs. 11.4% actual increase


Total bank lending, 2003–2006

∆Log(Total bank lending)county,2003−2006 = α + γCountyBeta + ε

∆ Total bank lending

.25

.3.3

5.4

.45

.5


Slope = -0.796 (0.127)

⇒ As Fed tightened, counties more exposed to the deposits channel saw lessbank portfolio and total bank mortgage lending


Total bank lending, 2003–2006

∆ycounty ,2003−2006 = α + γCountyBeta + δXcounty + ε

∆ Bank lending

(1) (2)

County beta −0.368*** −0.267***(0.132) (0.132)


1. Total bank lending declines by less than portfolio lending ⇒ compositionof bank lending shifts to PLS


Change in non-bank share, 2003–2006

∆Non-bank sharecounty ,2002−2006 = α + γCountyBeta + ε

Change in nonbank lending share

-.02

0.0

2.0

4.0

6


Slope = 0.208 (0.038)

⇒ Non-banks led the shift to PLS, gaining market share


Change in non-bank share, 2003–2006

∆ycounty ,2003−2006 = α + γCountyBeta + δXcounty + ε

∆ Nonbank lending share

(1) (1)

County beta 0.094** 0.124***(0.041) (0.040)


⇒ Non-bank share rose 12.4% more in a county with beta of 1 (maximallyexposed) than in a county with beta of 0 (unexposed).


Total mortgage lending (non-GSE), 2003–2006


∆ Total lending

(1) (2)

County beta −0.206* −0.085(0.116) (0.114)


1. Total lending is 8.5% lower (with controls) in a county with beta of 1(maximally exposed) relative to a county with beta of 0 (unexposed)

- controls for loan demand matter more for total lending

⇒ Implied aggregate contraction in total lending is only 4.52%

- due to substitution from bank portfolio lending to PLS lending


Substitution and aggregate impact

1. Use the cross-sectional coefficients to estimate the substitutionbetween bank portfolio (BP) and PLS lending.

Total lending TL = BP + PLS ⇒

−dPLS

dBP= −dTL− dBP

dBP=

(dTL/TL

dBP/BP× TL

BP− 1

)=−

(−0.085

−0.486× 1

1− 0.497− 1

)= 0.652

⇒ PLS offsets 65.2% of the contraction in bank portfolio lending

2. Similarly, non-bank lending substitutes 56.8% of the contraction inbank lending.

⇒ Impact of Fed tightening was substantially offset by PLS lending, ledby non-banks:

i bank portfolio lending fell by 25.9%

ii but total lending fell by only 4.52%

iii due to +16.8% PLS lending, led by +18.8% in non-bank lending


Financial fragility

1. Fed tightening induced shift from deposit-funded lending towholesale-funded PLS lending

2. PLS market has no government support, in contrast to GSE marketand bank portfolio mortgages

- GSE mortgages: (quasi-) government guarantee- bank portfolio mortgages: funded by government-insured deposits

⇒ PLS-funded mortgage market is much more exposed to runs/freezes

- such a run/freeze began in 2007 and only ended with governmentintervention


Takeaways

1. We analyze the impact of Fed tightening on mortgage lendingduring the housing boom through the lens of the deposits channel

2. We find that Fed tightening induced outflows of deposits and acontraction in bank portfolio mortgage lending

3. This contraction accelerated the shift to private-label securitization(PLS), led by non-banks, which largely undid the contractionaryimpact of Fed tightening

- investors’ newfound willingness to supply funding for PLS wasultimate driver of boom

4. Results closer to Bernanke’s (2010) view that tighter supervisionwould have been more effective than further raising rates


how monetary policy shaped the housing...

Documents